demo.py - output differs between downloaded model and detectron2 syntax

[horsto]

I am not sure I am understanding things correctly, but shouldn't these two commands yield the same output:

python ../demo/demo.py --config-file ../configs/COCO-InstanceSegmentation/mask_rcnn_R_101_FPN_3x.yaml \
 --video-input video-clip.mp4 --confidence-threshold 0.5 --output video-output3.mp4 \
 --opts MODEL.WEIGHTS detectron2://ImageNetPretrained/MSRA/R-101.pkl

python ../demo/demo.py --config-file ../configs/COCO-InstanceSegmentation/mask_rcnn_R_101_FPN_3x.yaml \
--video-input video-clip.mp4 --confidence-threshold 0.5 --output video-output3.mp4 \
--opts MODEL.WEIGHTS model_final_a3ec72.pkl

I am changing only the --opts MODEL.WEIGHTS to the downloaded version of the model (at least that's what I think I am doing).

In the first case I get a video that is devoid of any labeled overlay, and for the latter I get a video with lots of detected objects.